JPH07104938B2 - Method and device for creating dictionary pattern - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION A. Field of Industrial Application The present invention relates to a dictionary pattern creating method and apparatus,
It is useful when applied to pattern recognition of characters and figures.

B. Conventional Technology In pattern recognition of characters, figures, etc. in the conventional technology, a statistical method is used based on the characteristics of the input pattern to determine the similarity with the dictionary pattern and to associate it with the dictionary pattern. ing. That is, for example, the component feature points of the character "W" are taken as shown in FIG.
Each time, the input pattern (see FIG. 2A) is compared with the separately stored dictionary pattern information. In this comparison, feature points 1 to 5 (portions circled in the figure) are subdivided into square areas, and the state in which a part of the characters is present in a certain proportion or more in each area and the state in which it is not are respectively The similarity of the input pattern to the dictionary pattern is determined in correspondence with 1 "and" 0 ". More specifically, if the input pattern and the dictionary pattern match at positions corresponding to each other in the input pattern and the dictionary pattern, that is, if both are “1” or “0”, the input pattern is 0. And the dictionary pattern do not match, that is, when one is "1" and "0" and the other is "0" and "1", the sum is calculated for each of the feature points 1 to 5 as 1, The numerical value representing this sum is set by a threshold and evaluated. The smaller the numerical value at this time, the more the input pattern matches the dictionary pattern.

C. Problems to be Solved by the Invention As described above, in the related art, since two states of “1” and “0” are the basis of pattern recognition, in order to deal with the positional deviation and deformation of the input pattern, The dictionary pattern has a complicated structure, and this also has the drawback that the pattern recognition procedure is complicated.

SUMMARY OF THE INVENTION The present invention has been made in view of the above-described conventional art, and an object thereof is to provide a dictionary pattern creation method and apparatus for obtaining a dictionary pattern that can flexibly allow positional deviation and deformation of an input pattern with a simple configuration.

D. Means for Solving the Problems In the configuration of the method of the present invention for achieving the above object, the membership value representing the possibility that characteristic points of a pattern such as a character or a figure exist is subdivided in the x direction and the y direction. A step of creating a first fuzzy dictionary by determining for each feature point in the x direction and the y direction of the generalized standard pattern, and a membership value in the first fuzzy dictionary for each feature point and a membership value in the x direction. Comparing the membership values in the y direction for each element, selecting the smaller one of them to create the second ambiguous dictionary for the membership value of the element, and for each feature point of each element Compare membership values,
Among them, a step of selecting the largest value and setting it as a membership value for each element to create a dictionary pattern related to a standard pattern is characterized.

Further, the device of the present invention sets the membership value representing the possibility that the characteristic points of the pattern such as characters and figures exist in the x direction and the y direction of the standard pattern subdivided in the x direction and the y direction, as a characteristic value. First and second memory means for storing each point, and for each feature point, a membership value in the x direction in the first memory means and a membership value in the y direction in the second memory means for each element. First comparing means for selecting the smaller value among them, second storing means for storing the comparison result for each element of the first comparing means for each feature point, Based on the comparison result stored in the second storage means, the membership values of the respective feature points are compared for each element, and the second comparison means for selecting the largest value among them and the respective elements of the second comparison means. And having a third storage means for storing the comparison result for each cement as the membership value.

E. Operation According to the present invention having the above-mentioned structure, the membership values in the x-direction and the y-direction of each element by the first comparing means based on the first ambiguous dictionary stored in the first and second memory means. A second fuzzy dictionary for each feature point is created by comparing In addition, the maximum membership value of each element in the second fuzzy dictionary is selected by the second comparison means and stored in the third memory means.
The final dictionary pattern is formed in the memory means.

F. Example Hereinafter, an example of the present invention will be described in detail with reference to the drawings.

FIG. 2 (b) shows a standard pattern used to create a dictionary of the letter W. As shown in the figure, the present embodiment will be described by taking as an example the case of creating a dictionary pattern relating to the character W having the characteristic points 1 to 5.

An xy plane in which the horizontal direction of the drawing in FIG. 2B is the x direction and the vertical direction is the y direction is prepared, and a region (n × n (8 × 8 in this embodiment) of elements (on the xy plane)) is formed. (A region surrounded by the outermost contour of the square in FIG. 2B)
think of. Next, using the concept of fuzzy sets, the membership function when the feature points are distributed in the x direction is f
Let f _B (y) be the membership function when the characteristic points are distributed in the _A (x) and y directions.
Determine f _A (x) and f _B (y).

The character W having the characteristic points 1 to 5 as described above is n × n = 8 ×
Membership function f quantized by 8 elements
Table 1 shows _A (x) and f _B (y).

The numbers on the vertical axis (i) in the table indicate which of the characteristic points 1 to 5 they belong to. The numbers on the horizontal axis are the numbers of each element (x-direction address) of the one-dimensional standard pattern in the x direction at the feature point i in the case of the membership function f _A (x) and the membership function f _B (y). Represents the number (y-direction address) of each element of the one-dimensional standard pattern in the y-direction at the characteristic point i. The numerical values in Table 1 represent the membership value M (1 ≧ M ≧ 0) of each element. The membership value M is determined by the creator of the dictionary pattern from each characteristic point 1 to
5 is created by considering the area where they may exist and giving each element a numerical value of 1 ≧ M ≧ 0. In this case, the membership function f _A (x) for feature point 3
Is a membership value M of 0,0,0.3,0.8,1.0,0.8,0.3,0
Has been determined to have. Membership value M
= 1.0 means that the probability that the characteristic point 3 exists in the given area is the largest, and that the probability that the characteristic point 3 exists becomes smaller as the numerical value becomes smaller. Table 1 is called ambiguous dictionary I.

Membership function f _A (x) in the x direction for feature point 3
Membership functions f _B (y) in the y and y directions are shown in FIGS. 3 and 4, respectively.

Generally, the membership function f _A (x), f
As a result of giving _B (y), the fuzzy relation given by the following two equations holds for one feature point.

f _AB (x, y) = supMin [f _A (x), f _B (y)] (1) Equation (1) is calculated for each element in the membership functions f _A (x) and f _B (y). This means comparing membership values M and choosing the smaller one.

Specifically, the selection is made as follows. That is, as is clear from FIG. 3 and FIG. 4, for example, f _AB (0,0)
Selects “0” because the membership value of f _A (0) is “0” and the membership value of f _B (0) is “0”. Similarly, since f _AB (0,1) is f _A (0) = 0 and f _B (1) = 0, “0”, ... f _AB (3,0) is f _A (3) = 0.8, f
_{Since B} (0) = 0, “0”, f _AB (3,1) is f _A (3)
= 0.8, f _B (1) = 0, so “0”, f _AB (3,2) is f
_{A (3) = 0.8, f} B (2) = "0" for a 0, f _AB (3,
In 3), f _A (3) = 0.8 and f _B (3) = 0.2.
2 "and f _AB (3,4) are f _A (3) = 0.8 and f _B (4) = 0.7, so select" 0.7 "and repeat this selection up to f _AB (7,7).

The distribution of the feature points 1 to 5 in the dictionary space in which the Z-axis direction perpendicular to the drawing is added to the area shown in FIG. 2B is the space shown by the following union.

f _D (x, y) = Max [f _c1 (x, y), f _c2 (x, y), ... f _cm (x,
y)] (2) where f _ci (x, y) = f _{(AB) i} (x, y) (where i is the number of the feature point (see Table 1)), m = the number of feature points Note that f _{(AB) i} (x, y) is expressed for each feature point. I in this embodiment is 1 to 5.

The expression (2) means that the membership value M for each feature point obtained by the expression (1) is compared for each element, and the largest membership value M is selected as the membership value M for each element. There is.

Table 2 shows the case where the operation shown in the equation (1) is performed on the feature point 3, which is referred to as a fuzzy dictionary II.

The fuzzy dictionary II is also created for the feature points 1, 2, 4, and 5. Table 3 shows the operation performed in the formula (2). This represents the fuzzy space of all the characteristic points 1 to 5 of the character "W". That is, a dictionary pattern having a large membership value M in the vicinity of each of the feature points 1 to 5 is created.

FIG. 1 is a block diagram showing an apparatus for creating the dictionary pattern. As shown in the figure, input terminals 11,1
2 has a membership value (m in the x direction) of k feature points i.
Number) and the y-direction membership value (n number) via an input device (not shown). The memories 13 and 14 store the x-direction membership value and the y-direction membership value. Therefore, the memory 13 is m × k, and the memory 14 is
Has a storage capacity of n × k, and an ambiguous dictionary I corresponding to the above-mentioned Table 1 is formed, and x of the feature point i is
The membership value μ _xi (k) regarding the direction and the membership value μ _yi (l) regarding the y direction can be read out.

The x-address generator 15 and the y-direction address generator 16 calculate the x-direction address and the y-direction address of each element of a two-dimensional standard pattern (see FIG. 2B) in the x-direction and the y-direction for creating a dictionary. It occurs. That is, the output signals of the x-address generator 15 and the y-direction address generator 16 are information representing the x-direction address k (1≤k≤m) and the y-direction address l (1≤l≤m).

The comparator 17 compares the membership values μ _xi (k) and μ _yi (l) in the x-direction and the y-direction for each element, and outputs the smaller one. That is, if μ _xi (k) <μ _yi (l), then a _i = μ _xi (k), and if μ _xi (k) ≧ _yi (l)
Output a _i = μ _yi (l) and store it in the memory 18. Therefore, the vague dictionary II corresponding to the above-mentioned Table 2 is stored in the memory 18.
Is formed for each feature point i.

The comparator 19 stores the membership value a _i of each element stored in the memory 18 and the membership value W (i, J) (1) of each element stored in the memory 20 for each feature point i.
≤ i ≤ m, 1 ≤ j ≤ n), and the one with the larger value is output. That is, if W (k, l) <a _i, then W (k, l) = a _i ,
If W (k, l) ≧ a _i , W (k, l) = W (k, l) is output and stored in the memory 20. Therefore, the largest value of the membership values a _i in the fuzzy dictionary II formed for each feature point i is stored in the memory 20 at the address corresponding to each element, and the dictionary pattern corresponding to Table 3 above is stored. Is formed.

In the above embodiment, the standard pattern of the letter "W" is used to form the dictionary pattern. However, based on the standard pattern, arbitrary characters, figures, etc. can be input through the input terminals 11 and 12 of the apparatus shown in FIG. A dictionary pattern of arbitrary characters, figures, etc. can be freely created by appropriately inputting the X-direction and Y-direction membership values of.

H. Effect of the Invention As described in detail with the embodiments above, according to the present invention, an area that can exist for one feature point can be defined. Deformation is acceptable. At this time, the possibility that feature points exist is represented by the membership value M (0 ≦ M ≦ 1), and the size can be set intuitively or intuitively. Further, there is an effect that even if the number of characteristic points increases, they can be easily added.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 (a) is an explanatory view showing an input pattern, and FIG. 2 (b) is used for creating a dictionary. Explanatory drawing showing a standard pattern,
3 and 4 are graphs showing specific examples of membership functions. In the drawing, 13,14,18,20 are memories and 17,19 are comparators.

Claims (2)

[Claims] 1. A membership value representing the possibility that feature points of a pattern such as a character or a figure exist, for each feature point in the x and y directions of a standard pattern subdivided in the x and y directions. The step of determining and creating the first fuzzy dictionary, and comparing the membership value in the x direction and the membership value in the y direction in the first fuzzy dictionary for each feature point, for each element, Comparing the membership value of each feature point for each element with the step of creating a second fuzzy dictionary that selects the one value as the membership value of the element,
A method for creating a dictionary pattern, which comprises the step of selecting a largest value among them and using it as a membership value for each element to create a dictionary pattern related to a standard pattern. 2. A membership value, which indicates the possibility that feature points of a pattern such as a character or a figure exist, for each of the feature points in the x direction and the y direction of a standard pattern subdivided in the x direction and the y direction. First and second memory means for respectively storing the membership values in the x direction in the first memory means and membership values in the y direction in the second memory means for each element. First comparing means for comparing and selecting the smaller value among them; second storing means for storing the comparison result of each element of the first comparing means for each feature point; Based on the comparison result stored in the storage means, the membership values of the respective feature points are compared for each element, and the second comparison means for selecting the largest value among them and each element of the second comparison means are selected. A dictionary pattern creating apparatus comprising: a third storage unit that stores a comparison result as a membership value.